There's very little visibility into the aggregate behavior and operation of index builds today it's very difficult to answer across a large number of clusters:

• What is the normalized throughput of an index build
• What is the success rate of index builds
• What is the typical duration of each phase of an index build
• What phase of index builds do most failures occur
• How many indexes are built on a given day

To assist in answering these questions and others we should capture relevant metrics. We can take inspiration from resharding metrics.

When complete it should be possible to build a funnel chart/diagram detailing clusters progress through the end-to-end index build process and charts detailing the performance of index builds.